Llm Ie
Python toolkit for LLM-based NER, attribute, and relation extraction with a no-code web app.
Last verified:
What is Llm Ie?
Llm Ie is a comprehensive Python toolkit of building blocks for LLM-based named entity recognition (NER), attribute extraction, and relation extraction (RE). It turns unstructured text into structured data via a Prompt Editor agent for prompt engineering, multi-level document and sentence extraction, concurrent processing 3-5x faster, built-in network visualization, and a drag-and-drop web app — engine-agnostic across OpenAI, Ollama, vLLM, Huggingface, LiteLLM, and llama-cpp-python, reaching 92.3% F1 on biomedical information extraction.
Llm Ie pricing
Pricing model: Freemium
LLM-IE is completely free and open source. The Python package is available on PyPI with no paid tiers or subscription plans. Users only pay for the LLM API costs if using cloud-based LLM services like OpenAI API, OpenRouter, or Azure OpenAI. Local deployment with Ollama, vLLM, or Huggingface is free except for GPU/compute costs. The web application is available as a Docker image on Docker Hub for free local deployment.
Llm Ie pros
- Open source and free to use on PyPI
- Supports 6 major LLM platforms including OpenAI, Ollama, vLLM, Huggingface
- 3-5x faster concurrent processing in v0.4.0+
- 92.3% F1-score accuracy on biomedical extraction tasks
- Built-in web application with drag-and-drop no-code interface
- Interactive prompt editor with LLM agent for custom prompt templates
- Separate attribute extractor for complicated attribute schemas
- Async extraction methods for AsyncIO pipeline integration
- Built-in entity and relation visualization tools
- Customizable extraction granularity (sentence-level, document-level)
- Fuzzy matching with 93% Jaccard similarity threshold
- ±2 sentence context window awareness
- Supports reasoning models like o3 and Qwen3 in v1.1.0
- Global API rate limiting with max concurrency control
- Modular design with separable chunking and prompting methods
- Document object with span validation to prevent duplicate extractions
- Stream extraction mode for real-time processing feedback
Llm Ie cons
- Requires separate installation of LLM inference engines
- No built-in LLM engine installation or checking
- LlamaCppInferenceEngine deprecated in v1.4.0
- Performance varies depending on chosen LLM and quantization
- Batch processing issues fixed but existed in earlier versions
- Requires environmental variables for API keys setup
- Web application runs on fixed port 5000 by default
- Primarily optimized for biomedical domain extraction
Frequently asked questions about Llm Ie
What is LLM-IE?
LLM-IE is a comprehensive Python toolkit that provides building blocks for LLM-based named entity recognition, attribute extraction, and relation extraction pipelines. It transforms unstructured text into structured data using large language models with a workflow including prompt engineering, extraction, and visualization.
How do I install LLM-IE?
Install the Python package via pip: pip install llm-ie. Note that you must separately install at least one LLM inference engine (LiteLLM, Llama-cpp-python, Ollama, Huggingface_hub, OpenAI API, or vLLM) as the package does not install them automatically.
Which LLM platforms does LLM-IE support?
LLM-IE supports 6 major LLM platforms: OpenAI API, Ollama, vLLM, Huggingface_hub, LiteLLM, and Llama-cpp-python. OpenAI-compatible services like OpenRouter are also supported through OpenAIInferenceEngine with custom base_url.
What extraction types does LLM-IE support?
LLM-IE supports Named Entity Recognition (NER) with customizable granularity, Entity Attributes Extraction with flexible formats, and Relation Extraction (RE) for both binary and multiclass relations.
How does concurrent processing work?
Concurrent processing uses semaphore to better utilize computation resources, achieving 3-5x faster analysis in v0.4.0+. Set concurrent=True in extract_frames() for asynchronous prompting, or use the new async extract methods in v1.4.0+.
What is the Prompt Editor?
The Prompt Editor is an LLM agent that helps draft prompt templates following the schema required by extractors. You chat with it interactively via editor.chat() and after a few rounds it produces a prompt template to start extraction.
How do I visualize extraction results?
Use doc.viz_serve() to start a Flask app at port 5000 (default) for built-in entity and relation visualization. The web application also provides streaming frame extraction and download outputs with a drag-and-drop interface.
What chunkers are available?
LLM-IE provides UnitChunker classes like SentenceUnitChunker for sentence-by-sentence prompting, and ContextChunker classes like SlideWindowContextChunker that provide context (e.g., ±2 sentences) for better extraction accuracy.
What is the AttributeExtractor for?
The AttributeExtractor (added in v1.2.0) offloads complicated attribute extraction tasks from the FrameExtractor. For use cases requiring many attributes (like clinical notes with drug names, strength, dosage, frequency, route), it improves accuracy by dividing extraction tasks.
Is there a web application for no-code access?
Yes, LLM-IE has a drag-and-drop web application for no-code access. Pull the Docker image from Docker Hub with docker pull daviden1013/llm-ie-web-app:latest and run with docker run -p 5000:5000 daviden1013/llm-ie-web-app:latest. It includes interface for chatting with Prompt Editor and streaming frame extraction.